I have exhausted all my efforts trying to convert the Yolov11n_seg model into a HEF file for inference on Hailo8L, but I have been unable to succeed. I am in need of the correct YAML configuration file to proceed with the compilation.
Converting a completely new model for use on a Hailo device can be challenging and usually requires some familiarity with the Hailo toolchain. To better understand how to guide you, could you share a bit more about your background?
Have you already worked through the tutorials included in the Hailo AI Software Suite Docker?
How familiar are you with the standard model-conversion flow (as opposed to the Model Zoo flow)?
What steps have you taken so far in the conversion process?
For new or custom models, make sure you’re using the default model-conversion flow. The Model Zoo flow is designed specifically for the pretrained models that Hailo provides and isn’t suitable for converting brand-new architectures.
If your goal is mainly to get up and running quickly, you might find it easier to start with one of the models from the Model Zoo e.g. yolov8n-seg before moving on to something more recent like yolov11n-seg. Beginning with a supported model can help you understand the workflow and tooling, and then you can apply that experience when you move on to newer architectures.
If you can share what you’re trying to achieve with yolov11n-seg, we can help identify whether there’s a simpler path or confirm that the newer model is necessary for your use case.
I have successfully compiled the YOLOv8-seg model on Hailo, and everything runs smoothly with the default configuration. The entire process was fully automated — I didn’t need to modify anything, and it worked perfectly.
But that’s exactly where the problem begins: YOLOv11-seg is supposed to be supported according to the official statements, yet there is no complete or reliable guidance available. The information in the forum is extremely fragmented and inconsistent, and different staff members provide contradictory answers.
Without an official default configuration or a clearly validated reference, the compilation process becomes unnecessarily difficult. I really want to focus my time and energy on training the model itself, not fighting with the compilation pipeline. A proper official configuration or end-to-end workflow would help a lot.